What Is The Clock Frequency Of The Pandora ?

[StealthBagel]

On the one hand, it makes it seem pretty odd that Intel abandoned XScale. Do people want x86 that badly in their handhelds?

I think they need to go in a new direction, XScale was a pretty poor architecture.... every device I've had with it performed the same or worse than Samsung S3C chipsets at half the operating frequency (in PDAs)... they seemed to have some extensions for decoding video (like SIMD instructions I suppose) but either they weren't supported in any video app I tried (core media player, WMP..) or they just didn't help substantially, because I'll be damned if my 324 MHz S3C-based HP PocketPC doesn't achieve higher FPS on the same videos than my 624 MHZ Xscale based Dell PocketPC did. Anecdotal yes but I've had quite a number of PDAs and the results were mostly universal..

Also XScale handhelds I had got very poor battery life compared (the S3C based ones an order of magnitude more with the same LCD brightness)... and got toasty. Made my hand sweat even. The aluminum case on one didn't help. Whole thing was like a heatsink...

 

[Exophase]

On the one hand, it makes it seem pretty odd that Intel abandoned XScale. Do people want x86 that badly in their handhelds?

I think they need to go in a new direction, XScale was a pretty poor architecture.... every device I've had with it performed the same or worse than Samsung S3C chipsets at half the operating frequency (in PDAs)... they seemed to have some extensions for decoding video (like SIMD instructions I suppose) but either they weren't supported in any video app I tried (core media player, WMP..) or they just didn't help substantially, because I'll be damned if my 324 MHz S3C-based HP PocketPC doesn't achieve higher FPS on the same videos than my 624 MHZ Xscale based Dell PocketPC did. Anecdotal yes but I've had quite a number of PDAs and the results were mostly universal..

Also XScale handhelds I had got very poor battery life compared (the S3C based ones an order of magnitude more with the same LCD brightness)... and got toasty. Made my hand sweat even. The aluminum case on one didn't help. Whole thing was like a heatsink...


Blame the rest of the architecture and the programmers, not the CPU. In your case it sounds like it was being limited by bus speed. The CPU itself has a few problems (mainly higher load-use penalty which especially hurt the block transfer instructions) but for the most part it performed as well or better than its counterparts. If you don't believe me then look at bitbank's blog, he has no bad things to say about the XScale vs say, various OMAPs.

I was more wondering why Intel abandoned ARM in general than their specific XScale configuration.

 

[sparr]

That kind of information is under NDA

But we can say that recent tests make the Pandora roughly about 3 times faster than a bog standard gp2x when executing the same code (Not code compiled specifically for the Pandora).

60 MPH is "2 times faster" (200% faster) than 20MPH. 20+20*2=60.

Now, with that in mind, did you mean to say "3 times faster" (which means the same thing as "300% faster" or "4 times as fast") or "3 times as fast"?

 

[StealthBagel]

Blame the rest of the architecture and the programmers, not the CPU. In your case it sounds like it was being limited by bus speed. The CPU itself has a few problems (mainly higher load-use penalty which especially hurt the block transfer instructions) but for the most part it performed as well or better than its counterparts. If you don't believe me then look at bitbank's blog, he has no bad things to say about the XScale vs say, various OMAPs.

I was more wondering why Intel abandoned ARM in general than their specific XScale configuration.

Well if the CPUs are all embedded in SOCs you can't really take them out to compare, so if there was some major design mistake there, and a 624 Mhz XScale based chip is met or exceeded in performance by a 312 MHz S3C it's not going to matter to the consumer if the chipset implementation or the CPU itself is the problem.... at any rate Intel's SOCs paint a very bad picture for XScale. Maybe some other component needs to run at such high frequency to perform properly, but the CPU is internally divided down to normal speeds like what an S3C runs at.... come to think of it when I overclocked my 624 MHz XScale-based Axim to over 900 Mhz it performed the same or worse as at 624. When I overclock my S3C-based HP to 400 MHz+ performance rises dramatically. Something is seriously amiss.

 

[Eolair]

That kind of information is under NDA

But we can say that recent tests make the Pandora roughly about 3 times faster than a bog standard gp2x when executing the same code (Not code compiled specifically for the Pandora).

60 MPH is "2 times faster" (200% faster) than 20MPH. 20+20*2=60.

Now, with that in mind, did you mean to say "3 times faster" (which means the same thing as "300% faster" or "4 times as fast") or "3 times as fast"?

Where did you learn semantics like that?

Correct me if I'm wrong but....

First, you really don't say "1 times faster" or "1 times as fast".
Because "1 times" would mean 100% which is 0% faster (which isn't faster at all).

What you say is,
"Twice as fast" which equals 200% or "100% faster".

To continue,
"3 times faster" equals 300% or "200% faster".

And so on.

 

[Squidge]

Wow, some people looks at things in a totally different light than yourself. To me "1x faster" is none-existent, although to some people that would mean "twice as fast" (as technically it's the same speed as the original faster).

So, it's 2 times faster (ie, the original speed, plus 2 * the original speed).

Wow, I'm confusing myself now.

Right, to prevent any further confusion:

gp2x = 46 fps
pandora = 138 fps

and make up your own "x amount faster" stuff

 

[Pickle]

Wow, some people looks at things in a totally different light than yourself. To me "1x faster" is none-existent, although to some people that would mean "twice as fast" (as technically it's the same speed as the original faster).

So, it's 2 times faster (ie, the original speed, plus 2 * the original speed).

Wow, I'm confusing myself now.

Right, to prevent any further confusion:

gp2x = 46 fps
pandora = 138 fps

and make up your own "x amount faster" stuff

46 < 138 == SNES nirvana ? poopmypants : crylike baby

 

[Eolair]

Wow, some people looks at things in a totally different light than yourself. To me "1x faster" is none-existent, although to some people that would mean "twice as fast" (as technically it's the same speed as the original faster).

So, it's 2 times faster (ie, the original speed, plus 2 * the original speed).

Wow, I'm confusing myself now.

Right, to prevent any further confusion:

gp2x = 46 fps
pandora = 138 fps

and make up your own "x amount faster" stuff

Yeah, like you said in the first place, that's 3 times faster (or 200% faster )

(You can tell we have little to do here :rolleyes: )

 

[fusion_power]

gp2x = 46 fps
pandora = 138 fps

At the same clock speed?

 

[Exophase]

Blame the rest of the architecture and the programmers, not the CPU. In your case it sounds like it was being limited by bus speed. The CPU itself has a few problems (mainly higher load-use penalty which especially hurt the block transfer instructions) but for the most part it performed as well or better than its counterparts. If you don't believe me then look at bitbank's blog, he has no bad things to say about the XScale vs say, various OMAPs.

I was more wondering why Intel abandoned ARM in general than their specific XScale configuration.

Well if the CPUs are all embedded in SOCs you can't really take them out to compare, so if there was some major design mistake there, and a 624 Mhz XScale based chip is met or exceeded in performance by a 312 MHz S3C it's not going to matter to the consumer if the chipset implementation or the CPU itself is the problem.... at any rate Intel's SOCs paint a very bad picture for XScale. Maybe some other component needs to run at such high frequency to perform properly, but the CPU is internally divided down to normal speeds like what an S3C runs at.... come to think of it when I overclocked my 624 MHz XScale-based Axim to over 900 Mhz it performed the same or worse as at 624. When I overclock my S3C-based HP to 400 MHz+ performance rises dramatically. Something is seriously amiss.


You're arguing with me when all I said is that I don't know why Intel abandoned XScale (the CPU...) if they were so interested in mobile. The CPU is definitely running at a real 624MHz. Memory bandwidth is a significant bottleneck with video decoding, since it's a streaming application that doesn't cache well, and the software probably wasn't tuned at all for XScale. Microsoft in particular showed quite a lot of indifference towards doing this. There's really little question that XScale would perform better at other more CPU bound tasks. When you overclocked the CPU that high you probably had to shift down the bus speed (to maintain a higher CPU multiplier without making the overall frequency too high), resulting in worse RAM speed which runs at only one half bus speed.

The latest XScale SoCs Intel was working on were supposed to have L2 cache and a lot of fast scratchpad RAM (768KB), which I'm sure would have alleviated things a lot.

 

[Squidge]

gp2x = 46 fps
pandora = 138 fps

At the same clock speed?


I can't disclose that, but I think you can guess the answer

 

[Peter R]

No?

 

[javaJake]

I'm guessing "No". 138 FPS sounds... big...

 

[Yrx]

I'm guessing "No". 138 FPS sounds... big...

It sounds like death. Try playing a game at almost 3 times the normal speed. lol

 

[icurafu]

Squidge of all people should know that you can't guess a MHz based on a FPS performance percentage on two different SoCs.

But it looks like he is hinting at 3 x GP2X. Which is 3 x 200MHz. This was probably taken using SquidgeSnes. You could probably try adn get the same FPS at 200MHz and then over clock at a few intervals until you get the trend. From there you can determine the clock needed if clocked as a GP2X.

 

[StealthBagel]

You're arguing with me when all I said is that I don't know why Intel abandoned XScale (the CPU...) if they were so interested in mobile.

Calm down, I'm not arguing with you.... I just said Xscale isn't god's gift to the mobile world and other ARM-based architectures seem to have the upper hand right now. Don't be so defensive. Are you looking for an argument?

 

[Squidge]

But it looks like he is hinting at 3 x GP2X. Which is 3 x 200MHz. This was probably taken using SquidgeSnes.

Actually, it was taken using fba2x (my original release, not any of the updated ones)

 

[icurafu]

Woah! 138fps on finalburn is impressive.

 

[Squidge]

Yup, you can run benchmarks on systems until the cows come home, but since this machine is going to be running a lot of Emu's and not benchmarks, it's always best to actually use them instead when comparing performance.

(and no, the clock speed was not running at 600Mhz (or higher) for the tests)

 

[icurafu]

You already said MW was wrong with 550MHz, so I would guess 500MHz just to be even.